The 43rd Social Simulation & Service System Symposium
開催地
和文:
英文:
アブストラクト
In this paper we tried to investigate cooperative behavior of agents in an environment which includes agents with different types of learning algorithm (heterogeneous agents). Four types of learning algorithm will be used (e.g. Q-Learning, Q- SARSA Learning, PHC Learning, and WOLF-PHC Learning) to investigate cooperative behavior of agents. Q-Learning and Q-SARSA Learning has been used widely in multi-agent systems, but converges only to pure strategies. On the other hand, PHC Learning and WOLF-PHC Learning can learn mixed strategies and WOLF-PHC Learning is guaranteed to converge to Nash equilibrium of the repeated game. However, the algorithms work among homogeneous learners (self-play). We use a social dilemma game to illustrate the environment in which cooperative behavior among agents can be observed. The purpose of this investigation is to find a combination of learning types that can establish cooperative behavior among agents. To simplify this search process, we build a mechanism that can find a combination of learning that establishes cooperative behavior among agents.
各種検索
サポート
ヘルプ